VSE/ESA Performance when running under VM/ESA
From 20 October 1997:
The following is a summary of VSE guest performance for the two workloads we measure. In additional are a list of comments from which to help qualify the "it depends" statement that should always be included with this material. The majority of this information was taken from the VM/ESA Release 2.2 Performance Report GC24-5673-01 or the VM/ESA 2.1.0 performance report GC24-5801-00.
- ITR
- Internal Throughput Rate = measure of work per CPU second
- ETR
- External Throughput Rate = measure of work per Wallclock second
- %VM
- percentage of total processor time used by VM/ESA Control Program
DYNAPACE Workload
Environment | ITR | ETR |
%VM | Pct Nat ITR | Pct Nat ETR |
Runid |
||
|---|---|---|---|---|---|---|---|---|
| Native | 18.66 | 7.63 | 0% | 100% | 100% | L1NATxxx | ||
| V=R Dedicate | 16.48 | 7.57 | 8% | 88% | 99% | L1R88PF0 | ||
| V=V MDCache | 13.22 | 12.51 | 26% | 71% | 164% | L1V88PF0 | ||
Environment | ITR | ETR |
%VM | Pct Nat ITR | Pct Nat ETR |
Runid |
||
|---|---|---|---|---|---|---|---|---|
| Native | 31.18 | 7.88 | 0% | 100% | 100% | L2NATxxx | ||
| V=R Dedicate | 28.89 | 7.49 | 10% | 93% | 95% | L2R88PF1 | ||
| V=V MDCache | 20.25 | 12.33 | 28% | 65% | 156% | L2V88PF3 | ||
PACEX Workload
Environment | ITR | ETR |
%VM | Pct Nat ITR | Pct Nat ETR |
Runid |
||
|---|---|---|---|---|---|---|---|---|
| Native | 35.46 | 6.16 | 0% | 100% | 100% | LBNAT132 | ||
| V=R Dedicate | 25.38 | 5.94 | 16% | 72% | 96% | LB78REX1 | ||
| V=R | 19.14 | 5.98 | 34% | 54% | 97% | LB78REX6 | ||
| V=R MDCache | 18.93 | 9.31 | 36% | 53% | 151% | LB78REX2 | ||
| V=V | 18.24 | 6.13 | 37% | 51% | 100% | LB78VEXK | ||
| V=V MDCache | 19.21 | 10.70 | 36% | 54% | 174% | LB78VEXN | ||
| V=V MDC+VDisk | 19.83 | 16.15 | 35% | 56% | 262% | LB78VEY1 | ||
VSECICS Workload
Environment | ITR | ETR |
%VM | Resp Time |
Pct Nat ITR | Pct Nat ETR |
Runid |
||
|---|---|---|---|---|---|---|---|---|---|
| Native | 77.91 | 0% | 0.196 | 100% | NAT3208A | ||||
| V=R Dedicate | 72.93 | 65.76 | 3% | 0.226 | 94% | L1R88C90 | |||
| V=V MDCache | 65.32 | 60.21 | 12% | 0.237 | 84% | L1V88C90 | |||
Environment | ITR | ETR |
%VM | Resp Time |
Pct Nat ITR | Pct Nat ETR |
Runid |
||
|---|---|---|---|---|---|---|---|---|---|
| Native | 87.95 | 60.81 | 0% | 0.210 | 100% | 100% | NAT3207B | ||
| V=R Dedicate | 80.93 | 57.64 | 5% | 0.241 | 92% | 95% | VR13207C | ||
| V=V | 69.56 | 49.51 | 14% | 0.202 | 79% | 81% | VV13207K | ||
| V=V MDCache | 69.64 | 48.24 | 15% | 0.190 | 79% | 79% | VV13207F | ||
Things it depends on
- What do you mean by performance? ITR? ETR? Response Time?
- Obviously, the workload as seen above is a factor. ETR focus for batch workloads and Response time focus for CICS workload.
- MDC results beneficial if enough storage and CPU capacity exist.
- Preferred guests require I/O Assist for high ITR values, sharing DASD can minimize this.
- Extra hypervisor layers can impact results. For example, with VM/ESA running in a logical partition, only 1 V=R machine is possible and I/O Assist is not possible.
- For environments that do not benefit from I/O Assist, the larger the percentage of I/Os that are DASD, the better ITR picture. This is due to fast CCW translation which is only valid for DASD I/O.
- The slower the real DASD I/O hardware, the better the ETR and response time numbers are since minidisk cache and virtual disk in storage minimize I/O elapsed time. Even with control unit cache, MDC is faster.
- I/O patterns. High lock file I/O activity will benefit more from virtual disk in storage. Higher data re-references will show benefits from minidisk cache. Read-once data patterns will not see benefit.
- Overall system utilization. VM/ESA has background work such as monitor, MDC arbiter, accounting, etc. that occurs that is not proportional to activity on the system. This 'overhead' is relatively constant so when spread across more transactions, the base overhead becomes smaller. Several of the benchmark measurements above were made with low system utilization (less 70% CPU).
Update, January 7, 2005
z/VM 5.1.0 drops support for preferred guests and for I/O Assist. This has implications for VSE workloads, especially in processor-constrained environments. For more information, visit this tip.
Back to the Performance Tips Page